app.py

import os
import gradio as gr
from langchain_community.document_loaders import PyPDFLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_groq import ChatGroq
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableLambda

# ── Global state ──────────────────────────────────────────────
vectorstore    = None
qa_chain       = None
retrieved_docs = {}

# ── Groq key from HF Secret ───────────────────────────────────
GROQ_API_KEY = os.environ.get("GROQ_API_KEY", "")

# ── Embedding model (loaded once) ─────────────────────────────
embeddings = HuggingFaceEmbeddings(
    model_name="sentence-transformers/all-MiniLM-L6-v2",
    model_kwargs={"device": "cpu"},
    encode_kwargs={"normalize_embeddings": True}
)

# ── Helpers ───────────────────────────────────────────────────
def format_docs(docs):
    return "\n\n".join(
        f"[Page {doc.metadata.get('page', '?') + 1}]\n{doc.page_content}"
        for doc in docs
    )

def build_chain():
    llm = ChatGroq(
        api_key=GROQ_API_KEY,
        model="llama-3.3-70b-versatile",
        temperature=0.2,
        max_tokens=1024,
    )

    PROMPT = PromptTemplate(
        template="""You are a helpful assistant. Use the context below to answer the question.
If the answer is not in the context, say "I don't have enough information to answer that."

Context:
{context}

Question: {question}

Answer:""",
        input_variables=["context", "question"]
    )

    retriever = vectorstore.as_retriever(
        search_type="similarity",
        search_kwargs={"k": 4}
    )

    def retrieve_and_format(input_dict):
        question = input_dict["question"]
        docs = retriever.invoke(question)
        retrieved_docs["docs"] = docs
        return {
            "context": format_docs(docs),
            "question": question
        }

    return (
        RunnableLambda(retrieve_and_format)
        | PROMPT
        | llm
        | StrOutputParser()
    )

# ── Core functions ────────────────────────────────────────────
def process_pdf(pdf_file, progress=gr.Progress()):
    global vectorstore, qa_chain

    if pdf_file is None:
        return "⚠️ Please upload a PDF file."
    if not GROQ_API_KEY:
        return "❌ GROQ_API_KEY secret is not set in HF Space settings."

    try:
        progress(0.1, desc="Loading PDF...")
        loader = PyPDFLoader(pdf_file.name)
        pages = loader.load()

        progress(0.3, desc="Splitting into chunks...")
        splitter = RecursiveCharacterTextSplitter(
            chunk_size=800,
            chunk_overlap=100,
            separators=["\n\n", "\n", ".", " "]
        )
        chunks = splitter.split_documents(pages)

        progress(0.6, desc="Building FAISS index...")
        vectorstore = FAISS.from_documents(chunks, embeddings)

        progress(0.9, desc="Setting up RAG chain...")
        qa_chain = build_chain()

        progress(1.0, desc="Done!")
        return f"✅ Ready! Loaded **{len(pages)} pages** → **{len(chunks)} chunks**."

    except Exception as e:
        return f"❌ Error: {str(e)}"


def answer_question(question, history):
    if vectorstore is None or qa_chain is None:
        history.append({"role": "user",      "content": question})
        history.append({"role": "assistant", "content": "⚠️ Please upload a PDF first."})
        return "", history
    if not question.strip():
        return "", history

    try:
        answer = qa_chain.invoke({"question": question})

        docs = retrieved_docs.get("docs", [])
        if docs:
            pages = sorted(set(
                doc.metadata.get("page", 0) + 1
                for doc in docs
            ))
            answer += f"\n\n📄 *Sources: pages {pages}*"

    except Exception as e:
        answer = f"❌ Error: {str(e)}"

    history.append({"role": "user",      "content": question})
    history.append({"role": "assistant", "content": answer})
    return "", history


def clear_all():
    global vectorstore, qa_chain, retrieved_docs
    vectorstore    = None
    qa_chain       = None
    retrieved_docs = {}
    return [], "", "🗑️ Cleared. Upload a new PDF to start again."


# ── Gradio UI ─────────────────────────────────────────────────
with gr.Blocks(title="PDF RAG Chatbot") as demo:

    gr.Markdown("# 📚 PDF RAG Chatbot\nUpload a PDF and ask questions about it.")

    with gr.Row():

        # ── Left panel ────────────────────────────────────────
        with gr.Column(scale=1):
            gr.Markdown("### 📄 Upload Document")
            pdf_upload  = gr.File(label="Choose PDF", file_types=[".pdf"])
            process_btn = gr.Button("📥 Process PDF", variant="primary")
            status_box  = gr.Markdown("*Upload a PDF to begin.*")

        # ── Right panel ───────────────────────────────────────
        with gr.Column(scale=2):
            gr.Markdown("### 💬 Chat")
            chatbot = gr.Chatbot(height=500)               # ← type="messages" removed
            with gr.Row():
                question_box = gr.Textbox(
                    placeholder="Ask a question about your PDF...",
                    show_label=False,
                    scale=4
                )
                submit_btn = gr.Button("Send", variant="primary", scale=1)
            clear_btn = gr.Button("🗑️ Clear Chat & Reset")

    # ── Event handlers ────────────────────────────────────────
    process_btn.click(
        process_pdf,
        inputs=[pdf_upload],
        outputs=[status_box]
    )
    submit_btn.click(
        answer_question,
        inputs=[question_box, chatbot],
        outputs=[question_box, chatbot]
    )
    question_box.submit(
        answer_question,
        inputs=[question_box, chatbot],
        outputs=[question_box, chatbot]
    )
    clear_btn.click(
        clear_all,
        outputs=[chatbot, question_box, status_box]
    )

demo.launch(theme=gr.themes.Soft())